Michigan Math. J. 61 (2012), 151–160 Grimm’s Conjecture and Smooth Numbers Shanta Laishram & M. Ram Murty 1. Introduction In 1969, C. A. Grimm [8] proposed a seemingly innocent conjecture regard- ing prime factors of consecutive composite numbers. We begin by stating this conjecture. Let n ≥ 1 and k ≥ 1 be integers. Suppose n + 1, ..., n + k are all composite numbers; then there are distinct primes Pi such that Pi |(n + i) for 1 ≤ i ≤ k. That this is a difficult conjecture having several interesting consequences was first pointed out by Erd˝os and Selfridge [6]. For example, the conjecture implies there is a prime between two consecutive square numbers, something that is out of bounds even for the Riemann hypothesis. In this paper, we will pursue that theme. We will relate several results and conjectures regarding smooth numbers (defined in what follows) to Grimm’s conjecture. To begin, we say that Grimm’s conjecture holds for n and k if there are dis- tinct primes Pi such that Pi |(n + i) for 1 ≤ i ≤ k whenever n + 1, ..., n + k are all composites. For positive integers n>1 and k, we say that (n, k) has a prime representation if there are distinct primes P1, P2, ..., Pk with Pj |(n + j) for 1 ≤ j ≤ k. We define g(n) to be the maximum positive integer k such that (n, k) has a prime representation. It is an interesting problem to find the best possible upper and lower bounds for g(n). If n is the smallest prime greater than n, then Grimm’s conjecture implies that g(n)>n − n. However, it is clear that g(2m)<2m for m>3. The question of obtaining lower bounds for g(n) was attacked using methods from transcendental number theory by Ramachandra, Shorey, and Tijdeman [15], who derived n 3 g(n) ≥ c log log log n for n>3 and an absolute constant c>0. In other words, for any sufficiently large natural number n, (n, k) has a prime representation if k (log n/log log n)3. We prove the following theorem. α<1 g(n)<nα Theorem 1. (i) There exists an 2 such that for sufficiently large n. Received July 7, 2010. Revision received May 9, 2011. Research of the second author partially supported by an NSERC Discovery grant. 151 152 Shanta Laishram & M. Ram Murty (ii) For ε>0, we have |{n ≤ X : g(n) ≥ nε}| X exp(−(log X)1/3−ε ), where the implied constant depends only on ε. We show in Section 3 that 0.45 <α<0.46 is permissible in Theorem 1(i). For real x and y, let (x, y) denote the number of positive integers ≤ x all of whose prime factors do not exceed y. These are y-smooth numbers, which have been well studied. In 1930 Dickman [3] proved that, for any α ≤ 1, (x, x α) lim x→∞ x exists and equals ρ(1/α), where ρ(t)is defined for t ≥ 0 as the continuous solution of the equations ρ(t) = 1 for 0 ≤ t ≤ 1 and −tρ (t) = ρ(t − 1) for t ≥ 1. Later authors derived refined results; see [11] for an excellent survey on smooth num- bers. An important conjecture on smooth numbers in short intervals is as follows. Conjecture 1.1. Let ε>0. For sufficiently large x, we have (x + x ε, x ε ) − (x, x ε ) x ε. This remains an open question, but it leads to the following statement. Theorem 2. Assume Conjecture 1.1 holds, and let ε>0. Then g(n)<nε for large n. Let pi denote the ith prime. As a consequence of Theorem 2, we obtain the fol- lowing result. Corollary 1.2. Assume Grimm’s conjecture and Conjecture 1.1. Then, for any ε> 0, ε pi+1 − pi <pi (1) for sufficiently large i. If we assume Grimm’s conjecture alone, then Erd˝os and Selfridge [6] have shown that 1/2 pi+1 − pi (pi/log pi ) , a statement well beyond what the Riemann hypothesis would imply about gaps between consecutive primes. Indeed, the Riemann hypothesis implies an upper 1/2 bound of O(pi (log pi )). In 1936, Cramér [2] conjectured that 2 pi+1 − pi (log pi ) . If Cramér’s conjecture is true, then the result of [15] implies Grimm’s conjecture— at least for sufficiently large numbers. In [12], Laishram and Shorey verified 10 Grimm’s conjecture for all n<1.9 × 10 . They also checked that pi+1 − pi < 2 8 1 + (log pi ) for i ≤ 8.5 × 10 . It is worth mentioning the existence of several weaker versions of Grimm’s con- jecture that have also been attacked using methods of transcendental number the- ory. For an integer ν>1, we denote by ω(ν)the number of distinct prime divisors Grimm’s Conjecture and Smooth Numbers 153 ν ω( ) = . of and let 1 0 A weaker version of Grimm’s conjecture runs as follows: n + n + ... n + k ω k (n + i) ≥ k. If 1, 2, , are all composite numbers, then i=1 This conjecture, too, remains open, though much progress toward proving it has been made by Ramachandra, Shorey, and Tijdeman [16]. We define g1(n) to be the maximum positive integer k such that l ω (n + i) ≥ l i=1 for all 1 ≤ l ≤ k. Observe that g1(n) ≥ g(n). In Section 5 we prove the following. γ <γ < 1 Theorem 3. There exists a with 0 2 such that γ g(n) ≤ g1(n)<n (2) for large values of n. γ = 1 − 1 We show in Section 5 that 2 390 is permissible. This result will be proved as a consequence of our next theorem, which is of independent interest. Theorem 4. Suppose there exist 0 <α< 1 and δ>0 such that 2 m + mα m π − π ≥ δmα j j (3) j≤mα for large m. Then g (n)<nγ with 1 1 − δ(1 − α) 1 γ = max α, < 2 − δ 2 for large n. A conjecture arising from primes in short intervals states that x α π(x + x α) − π(x) ∼ as x →∞ log x (see e.g. [13]). Assuming this conjecture, for m →∞we obtain α α m α m + m m j m 1 π − π ∼ m = j j log log m j − log j j≤mα j≤mα j j≤mα 1 m log mα mα dt ∼ . log t log m 1 t − 1 log m u = log t Taking log m ,weget m + mα m π − π j j j≤mα α du ∼ mα = mα − ( − u) α =−mα ( − α) [ log 1 ]0 log 1 0 1 − u 154 Shanta Laishram & M. Ram Murty α1 as m →∞. Continuing as in the proof of Theorem 4, we obtain g1(n)<n with 1 + (1 − α)log(1 − α) α = max α, . 1 2 + log(1 − α) Since log(1 − α) ≈−α for 0 <α<1, it follows that − 1 + (1 − α)log(1 − α) 1 − α(1 − α) 1 α 1 ≈ = ( − α + α 2) − + ( − α) − α 1 1 2 log 1 2 2 2 1 α 1 ≈ (1 − α + α 2) 1 + = (2 − α + α 2 + α3) 2 2 4 and the function 1 (2 − α + α 2 + α3) attains its maximum at α = 1 , where the 4 3 γ value of α1 ≈ 0.4567. We are therefore unlikely to get a result with g1(n)<n α and γ<0.4567 by these methods. This value g1(n) = O(n ) does seem to agree with the permissible value of 0.45 <α<0.46 in g(n) = O(nα). γ<1 It was noted by Erd˝os and Selfridge in [6] that “the assertion 2 seems to follow from a recent result of Ramachandra [14] but we do not give the details here.” In [5], Erd˝os and Pomerance noted again that, “from the proof in [14], it follows that there is an α>0 such that for all large n a positive proportion of the integers in (n, n + nα] are divisible by a prime which exceeds n15/26. Using this result with the method in [6]givesg(n)<n1/2−c for some fixed c>0 and all large n.” However, there is no proof anywhere in the literature about this fact. We give a complete proof here by generalizing the result of [14] in Lemma 2.5. 2. Preliminaries and Lemmas We introduce some notation. We shall always write p for a prime number. Let r !(n) be the von Mangoldt function, which is defined as !(n) = log p if n = p for some positive integer r and as 0 otherwise. We write θ(x) = p≤x log p. For real x and y, let (x, y) denote the number of positive integers ≤ x all of whose prime factors do not exceed y. We also write log2 x for log log x. We begin with some results from prime number theory. Lemma 2.1. Let k, t ∈ Z and x ∈ R. Then: π(x) < x + 1.2762 x> ; (i) log x 1 log x for 1 (ii) pt >t(log t + log2 t − c1) for some c1 > 0 and for large t; (iii) θ(x) ≤ 1.00008x for x>0; θ(p )>t( t + t − c ) c > t; (iv) t √ log log2 2 for some 2 0 and for large and (v) k! > 2πke−kk ke1/(12 k+1) for k>1. The estimate (ii) is due to Rosser and Schoenfeld [18].